Symbolic motion analysis in digital subtraction angiography: a preliminary study

Abstract

displacements are quasi-homogeneous on a segment based analysis, a f i r s t approach to s tudy the twodimensional motions of the main coronary ar ter ies is propsed. A knowledge-based system extracts a n d interprets t h e d y n a m i c i n f o r m a t i o n . M o t i o n interpretations a r e obtained in two temporal image sequences a c q u i r e d in s t a n d a r d condi t ions, A prel iminary example of the system in te rpre ta t ion using declarative knowledge and d a t a representation is shown. T h e appl ica t ion of t h e homogeneous segment concept has permit ted to obtain a suitable arteries’ motion analysis. I . INTRODUCTION Visual interpretation includes perception, shape recognition, understanding, decision making and learning. Idcntification of useful entities in the image is one of the key problems to solve[l]. Interactions are simultaneous and varied, and a subjective inspection does not always differenciate between them. Knowledge based systems, aimed at the understanding of temporal image sequences, produce quantitative and qualitative evaluation of either objects or dynamic behaviors. Robot vision [ 2 ] and medical diagnosis (31, [4]. (51 are among the most important application fields. The objective in digital subtraction angiography is motivated by the need to work with qualitative, structural, morphological and kinetic properties. The analysis of cardiac movement is extremely complex. Normally, a highly qualified specialist is required to interpret these images. On the other hand, a great amount of the available angiography equipments take image sequences at one view angle and not at two angles simultaneously. Then, two dimensional motion analysis is rather used than three dimensional analysis. Most of the published works about the interpretation of cardiac angiographies are focused on the detection of lesions and do not include the temporal dimension. Usually, no more than two images are used. Studies are done with only two extreme positions of the heart, for example in ventricular analysis [6]. This work deals with knowledge based systems and image processing methods to analyse time-varying cardiovascular pictures. The temporal information given by the contrast medium circulation and the ECG signal is not considered in this work. Heart’s local and global behavior can be examined then, following only the coronary arteries’ displacements. This type of analysis relates three elements: artery’s position, magnitude and direction of its movement. The two-dimensional (2D) velocity vector associated to each artery’s point displacement is the basic data representation. The movement is qualitatively analysed in two temporal sequences recorded almost simultaneously with a biplane acquisition system R A 0 (right anterior oblique) and LAO (left anterior oblique) views. 11. METHOD OVERVIEW To achive qualitative kinetic description and motion analysis several steps must be accomplished: central line extraction, binary image representation and motion estimation. They are carried out by low level processings already reported in 171. Kinetics description is used to effectively recognize the variation of points’ position. Regions of motion are identified assuming that the objects’ movements are homogeneous on a segment based analysis. This concept guides the object recognition task to produce meaningful relationships between different image descriptions, created to represent the dynamic evolution of the scene. Namely, a set of intermediate descriptive elements must be available to go from calculated values to symbols. The context layers are structured in several levels: ( i ) . Velocity vector, artery’s segment angular orientation, average angle and magnitude of a selected vector field. (ii). A priori knowledge is applied to build the temporal concepts of expansion, contraction. discontinuities in the global movement direction, local and global tendencies. (iii). Considering a linear, discrete and convergent time, a verbal like temporal scale was designed to identify the heart cycles. Each vessel is an ordered and connected set of points in 2D. A homogeneous segment is defined as a group of adjacent points, whose angular orientation, with respect to the corresponding artery line segments, satisfies a previously stated inference. Motion interpretation’s labels of each branch, are assigned during the analysis of two consecutive images. The labels are added to a declarative knowledge base. They include the branch name, the view angle, the differential time, the interval’s length (if it is a branch segment) and the identified movement. In order to reduce the knowledge base size, i t is necessary to determine which segments are relevant. All segments. smaller than a pre-defined mesure, are not included in the knowledge base. This heuristic criterion is used to filter the original data and still we have the same global movement interpretations. Labels are found by means of a data directed process. Movement configurations are identified through goal directed reasoning. 0-780313771/93 $3.00 01993 JEEE 598 111. RESULTS AND DISCUSSION The first step was to itlciitify aid classify homogeneous Inoviiig segiiieiits. A Ilierarchy is bidt a i d the sclection criterion is npplicd. The system is 0l)eriited from a graphic interface where the user can select m y qiicstion related to the description of branch events, discontinuities in the global movement direction, cycles or certain configurations. For example at t=3, after the nppliczztion of the heuristic criterioil, the inter-ventricdar branch (AIV) is described as i‘ollows: seg(AlV. rao-30. 3. int(0.33). expa). seg(A1V. rao-30. 3, int(39. 71). expa). seg(AIV. rao-30. 3. int(72. 100). cont). seg(AI\’, rao-30, 3. int(l0l. 129). expa). seg(A1V. rao-30. 3 int(130. 178). cont). mov-global(A1V. rao-30, 3 expa). These statements indicate that the AIV has a global expansion inovernent. at t=3, in the RA0 view. This branch is fonned by 5 sigtuficative homogeneous seemeas, 3 with a local espansion tendency and two with a local contraction tendency. The other three branches. the circunlnex artery (CX). the lateral artery (L) and the diagonal artery (D) are described in the same way. seg(L. rao-30. 3, int(0. 17). expa). seg(L, rao-30, 3. int(29. 87). expa). mov-global(1,. rao-30. 3. expa). seg(D. rao-30, 3. int(4. 46). expa). seg(D. rao-30. 3. int(J7. 68), cont). seg(D, rao-30. 3. int(78. 150). cont) mov-global(D, rao-30. 3. comp). seg(CS. rao-30. 3. int(15, 107). expal. mov-global(CS. rao-30. 3. expa). All the detected segments in the mentioned arteries, are shown i n Figure l., where C stands for contraction and E for cxp”on. The rules identify certain coidgurations, according to simple principles. For instaice, the rnaxiinal expansion occurs only once in a cardiac cycle; this instant should be the same in [he two image sequences for all the branches whose movement is parallel to the image plane. The system identifies this instant and then proposes a ternpral description of the heart cycle for the given sequences. IV. CONCLUSION A first approach to syiubolic analysis of inotion in digital subtraction angiography has been presented. The solution considers the image sequence. the heart cycles aid is based on a temporal reasoiing to interpret some dynamic aspects 01 the heart. The expected contribution of the knowledge based s!’stenis, for the spatio-temporal interpretation of an mgiographic image sequence. is to serve as a support to identify abiionnal heart moveinents and its functional properties. A large aiiouiit of work i s still needed to denioiistrate the diagnostic value of this approach. Tllree-diniensional inotion remains to be studied as well. ACKNOWLEDGEMENT Tfiis work was supported in part by the FUNDAYXCUCHOCONICIT-CEFl (Venezuela France Cooperation Program). ..\RECOh? and Project E-08 (SIC) BID/COMCIT. L D AIV Figure 1. Systeiii’s iiitcrpretaiioii at 1=3 R E F E R E N C E S [ I ] J. L Coatrieux. M. Garreau, R. Collorec & G. Carrauli “Signal. iinage et iittelligetice nrtijcielie et1 me‘deciire“. Proceedings of the Intemational Conference ‘Les Entrctieiis de Lyon‘ on Medical Imagery and Expert Systems. March 1988. [2] E. D. Dickmans, B.tvlysliwetz & T. Christians. “Ail itiiegmrerl spatio-remporczl npprouch lo auloinutic visual guidnrrce oj aulonornus vehicles”. IEEE Transactions on Systems. X.lan and Cybernetics. Vol 10, No 6. pp. 1273-1284, Nov,/Dic 1990. [3] H. Nieniann, et al. “A kmrocvlcdgr Onred system for nilo/vsif O/ gated blood pool studies". EEE Transactions 011 Pattern Analysis and Machine Intelligence, Vol. PAMI-7. No 3. pp. 146-259. May 1985. sequences””. Pattern Recognition Letters 8. pp. 87102. Sept. 1988. represeiilatioii arid irilerpretntioii jor timevm-yiiig dara: I / I C ALVEN system”. Computational Intelligence I . pp. 16-32. 1985. Applicntioii ri In ddcoinpositioii du sigiral e‘lectromyograplrique et ci la recoiistriicrion et I’eriqrrernyc 30 de sfrucfures vasculaires”. These. Universite de Relines I , 1988. [7) S. Ruan. A. Bruno. R. Collorec & J.L. Coatrieun. “Esiittuiiiuti de inouvemerit 3 0 eii coromrogrczphir”. Actes 13 cme Colloque GRETSI, Juan les Pins, pp. 801-804, Sept. 1991. [J] G. Sagerer. “Autoinntic interpretuiion of medical irrulge [5] J. K. Tsotsos. “Knowlrdge orgoirizatioir atid ils role 111 [6] M. Garreau. “Sipial , image et inlelligettcc art@ielle.